Performance
Guide status: Measurement guidance. Last reviewed 14 August 2026. Metric availability, attribution rules and reporting windows must be confirmed in the installed version.
Purpose
Performance helps the team understand what was delivered, how customers interacted and what commercial outcomes were observed. Its role is to connect measurement to the decision and action that produced an experience, while showing the limits of what the evidence can prove.
A dashboard value is not meaningful without its definition, scope, window and state. Performance should not turn an association into a causal claim or substitute an estimate for an observed result.
Before you begin
- Confirm the store, market, currency, time zone and reporting window.
- Check that the intended placement was published and capable of exposure during that window.
- Understand the primary measure and guardrails chosen before the work began.
- Review visible data-freshness, processing and consent states.
Read the measurement chain
Begin with eligibility: the number or set of visits for which the experience could validly apply. Then consider delivery or exposure, followed by interaction and finally the defined commercial outcomes. Each stage can explain why a later value is low or unknown.
Use the correct denominator. For example, an interaction rate based on verified exposures answers a different question from one based on all sessions. Definitions should travel with the figure so that teams do not compare unlike measures.
Choose the right evidence standard
Descriptive reporting can show what happened during a period. A comparison can show that two states differed. Neither alone proves that a Liva 7 action caused the difference. When a decision requires causal confidence, use a controlled experiment where practical and ethical.
Read estimates, projections and modelled influence only as labelled planning evidence. Keep them visually and conceptually separate from observed orders, revenue or other realised outcomes.
Interpret responsibly
- Check sample size and duration before drawing a conclusion.
- Look for inventory, promotion, market, traffic or theme changes during the window.
- Review guardrails as well as the primary measure.
- Treat zero, Unknown, Learning and Unavailable as different states.
- Record when an attribution or identity limitation affects the result.
Safe use and fallback
If measurement is incomplete, do not manufacture an uplift, substitute a projection or silently change the window. Preserve the last valid state with its timestamp, or show Unknown or Learning. If a decision cannot wait, explain the missing evidence and use the lowest-risk reversible action.
You are finished when
- scope, definition, denominator, window and freshness are understood;
- observed and modelled measures are distinct;
- the conclusion matches the strength of the design; and
- the next action records both learning and limitations.